SETUP PACKAGES AND READ IN IMAGES¶

In [2]:
#import needed libraries
from matplotlib import pyplot as plt
import pandas as pd
import numpy as np
import tensorflow as tf
import tensorflow_hub as hub
import os
import cv2
import imghdr
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Flatten
from tensorflow.keras import layers

WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead.


C:\Users\eebla\AppData\Local\Temp\ipykernel_36408\1819497594.py:9: DeprecationWarning: 'imghdr' is deprecated and slated for removal in Python 3.13
  import imghdr
In [3]:
#getting rid of images that are not the correct extension or are corrupt.
# also manually got rid of small images <10KB
data_dir = (r'C:\Users\eebla\OneDrive\Documents\side_project')
image_exts = ['jpeg','jpg', 'bmp', 'png']
for image_class in os.listdir(data_dir): 
    for image in os.listdir(os.path.join(data_dir, image_class)): #it is looping through each of the folders and then all the images in each
        image_path = os.path.join(data_dir, image_class, image)
        try: 
            img = cv2.imread(image_path)
            tip = imghdr.what(image_path)
            if tip not in image_exts: 
                print('Image not in ext list {}'.format(image_path))
                os.remove(image_path)
        except Exception as e: 
            print('Issue with image {}'.format(image_path))

LOAD IN IMAGES WITH TENSORFLOW¶

In [5]:
#I dont have an NVIDIA graphis card:(
# fun fact if a filename is in chinese it breaks it
#defualt size is 256,256 and 32 per batch, shuffle is true going to need to get them to 224x224 for transfer learning model
#tf.keras.utils.image_dataset_from_directory??

#set wanted image size
IMAGE_SIZE=[224,224]
#set up the directory, tf likes to have subfolders with labels
data_dir = (r'C:\Users\eebla\OneDrive\Documents\side_project')
data= tf.keras.utils.image_dataset_from_directory(data_dir,image_size=(IMAGE_SIZE))

Found 552 files belonging to 2 classes.
In [6]:
#iterate through images and look at some of them 
#dachshunds are labeled as 1 and chihuahuas are labled as 0
# can run this a bunch and see different batches of 32
data_iterator = data.as_numpy_iterator() #images represented by numpy arrays, grabs batches of 32
batch = data_iterator.next()
batch[0] #this is the arrays reprensting the image
batch[1] #this array represents the labels aka dachshund or seal

fig, ax = plt.subplots(ncols=4, figsize=(26,26))
for idx, img in enumerate(batch[0][:4]):
    ax[idx].imshow(img.astype(int))
    ax[idx].title.set_text(batch[1][idx])
No description has been provided for this image

PREPROCESS¶

Need to scale data and then split into train, validation and test¶

In [8]:
# SCALE- need to scale the arrays from 0 to 1 instead of 0 to 255 (the color scale goes to 255), its in RGB with tensorflow, its BGR with open_cv.
#tensorflow data api -> can see what other functions you can do in the pipeline

#Need to apply this to the pipeline instead of doing it after batch comes in
#x is the images and y is the labels
data=data.map(lambda x,y:(x/255,y)) #the map function is from tf
batch=data.as_numpy_iterator().next()

#checking, looks good this will work. 
print(f'Min: {batch[0].min()}')
print(f'Max: {batch[0].max()}')

#they are all black
fig, ax = plt.subplots(ncols=4, figsize=(128,128))
for idx, img in enumerate(batch[0][:4]):
    ax[idx].imshow(img) #take out as type is int because they would all be 0 since we scaled it
    ax[idx].title.set_text(batch[1][idx])

Min: 0.0
Max: 1.0
No description has been provided for this image

SPLIT INTO TRAIN, VALIDATE AND TEST¶

In [10]:
print(len(data)) #there are 18 batches of 32

train_size = int(len(data)*.7) #12
val_size = int(len(data)*.2) #3
test_size = int(len(data)*.1) #1

print(train_size)
print(val_size)
print(test_size)

#the take is telling how many batches to do, data is shuffled so there are a mix of 0s and 1s in each batch
train = data.take(train_size) 
val = data.skip(train_size).take(val_size)  
test = data.skip(train_size+val_size).take(test_size) 

18
12
3
1

CNN with augmentation and transfer learning¶

In [12]:
import tf_keras as tfk
input_shape = (None,224, 224, 3)
hub_url = "https://www.kaggle.com/models/google/mobilenet-v2/TensorFlow2/035-224-feature-vector/2"

model = tfk.Sequential([
    tfk.layers.RandomFlip(mode='horizontal'),  #randomly flips horizontal
    tfk.layers.RandomRotation(0.1), #randomly rotates
    tfk.layers.RandomZoom(0.1),
    hub.KerasLayer(hub_url, trainable=False),  # Can be True, see below.
    tfk.layers.Flatten(),
    tfk.layers.Dense(1, activation='sigmoid') # output layer
])


# Compile the model
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
model.build(input_shape)
model.summary()

WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\backend.py:873: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.

WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tensorflow_hub\resolver.py:120: The name tf.gfile.MakeDirs is deprecated. Please use tf.io.gfile.makedirs instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tensorflow_hub\resolver.py:120: The name tf.gfile.MakeDirs is deprecated. Please use tf.io.gfile.makedirs instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tensorflow_hub\module_v2.py:126: The name tf.saved_model.load_v2 is deprecated. Please use tf.compat.v2.saved_model.load instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tensorflow_hub\module_v2.py:126: The name tf.saved_model.load_v2 is deprecated. Please use tf.compat.v2.saved_model.load instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\optimizers\__init__.py:317: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\optimizers\__init__.py:317: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.


Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 random_flip (RandomFlip)    (None, 224, 224, 3)       0         
                                                                 
 random_rotation (RandomRot  (None, 224, 224, 3)       0         
 ation)                                                          
                                                                 
 random_zoom (RandomZoom)    (None, 224, 224, 3)       0         
                                                                 
 keras_layer (KerasLayer)    (None, 1280)              410208    
                                                                 
 flatten (Flatten)           (None, 1280)              0         
                                                                 
 dense (Dense)               (None, 1)                 1281      
                                                                 
=================================================================
Total params: 411489 (1.57 MB)
Trainable params: 1281 (5.00 KB)
Non-trainable params: 410208 (1.56 MB)
_________________________________________________________________
In [13]:
#fitting model on training- validation went down around 10 then went back up
hist = model.fit(train, epochs=10, validation_data=val)

Epoch 1/10
WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\utils\tf_utils.py:492: The name tf.ragged.RaggedTensorValue is deprecated. Please use tf.compat.v1.ragged.RaggedTensorValue instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\utils\tf_utils.py:492: The name tf.ragged.RaggedTensorValue is deprecated. Please use tf.compat.v1.ragged.RaggedTensorValue instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\engine\base_layer_utils.py:384: The name tf.executing_eagerly_outside_functions is deprecated. Please use tf.compat.v1.executing_eagerly_outside_functions instead.


WARNING:tensorflow:From C:\Users\eebla\anaconda3\Lib\site-packages\tf_keras\src\engine\base_layer_utils.py:384: The name tf.executing_eagerly_outside_functions is deprecated. Please use tf.compat.v1.executing_eagerly_outside_functions instead.


12/12 [==============================] - 13s 475ms/step - loss: 0.5000 - accuracy: 0.7188 - val_loss: 0.3683 - val_accuracy: 0.7812
Epoch 2/10
12/12 [==============================] - 5s 368ms/step - loss: 0.2490 - accuracy: 0.8880 - val_loss: 0.1448 - val_accuracy: 0.9583
Epoch 3/10
12/12 [==============================] - 5s 406ms/step - loss: 0.1514 - accuracy: 0.9740 - val_loss: 0.1287 - val_accuracy: 0.9792
Epoch 4/10
12/12 [==============================] - 5s 401ms/step - loss: 0.1070 - accuracy: 0.9818 - val_loss: 0.0827 - val_accuracy: 0.9896
Epoch 5/10
12/12 [==============================] - 4s 348ms/step - loss: 0.0904 - accuracy: 0.9870 - val_loss: 0.0702 - val_accuracy: 1.0000
Epoch 6/10
12/12 [==============================] - 6s 472ms/step - loss: 0.0792 - accuracy: 0.9870 - val_loss: 0.0524 - val_accuracy: 1.0000
Epoch 7/10
12/12 [==============================] - 9s 722ms/step - loss: 0.0544 - accuracy: 0.9948 - val_loss: 0.0342 - val_accuracy: 1.0000
Epoch 8/10
12/12 [==============================] - 8s 642ms/step - loss: 0.0445 - accuracy: 0.9948 - val_loss: 0.0596 - val_accuracy: 0.9896
Epoch 9/10
12/12 [==============================] - 5s 421ms/step - loss: 0.0496 - accuracy: 0.9896 - val_loss: 0.0403 - val_accuracy: 0.9896
Epoch 10/10
12/12 [==============================] - 6s 503ms/step - loss: 0.0446 - accuracy: 0.9948 - val_loss: 0.0280 - val_accuracy: 1.0000

PLOT THE PERFORMANCE¶

In [15]:
#hist.history can see the accuracy, loss is stored for training and validation
#looks great, both the training and validation loss is decreasing over time
#plot loss
fig = plt.figure()
plt.plot(hist.history['loss'], color='teal', label='loss') #traiing loss
plt.plot(hist.history['val_loss'], color='orange', label='val_loss') #validation loss
fig.suptitle('Loss', fontsize=20)
plt.legend(loc="upper left")
plt.show()
No description has been provided for this image
In [16]:
#plot accuracy
fig = plt.figure()
plt.plot(hist.history['accuracy'], color='teal', label='accuracy') #training accuracy
plt.plot(hist.history['val_accuracy'], color='orange', label='val_accuracy') #validation accuracy
fig.suptitle('Accuracy', fontsize=20)
plt.legend(loc="upper left")
plt.show()
No description has been provided for this image

EVALUATION ON TEST SET¶

In [18]:
loss, acc = model.evaluate(test)
print(f"Test Loss: {loss}, Test Accuracy: {acc}")

1/1 [==============================] - 2s 2s/step - loss: 0.0169 - accuracy: 1.0000
Test Loss: 0.01686013862490654, Test Accuracy: 1.0

MORE testing¶

In [20]:
#test on my dachshunds Cannoli and Biscotti and my old boss's chihuahua peachy
#since we are using open cv its showing them in bgr that is why it looks a little off colorwise, fixed it
#Cannoli
img_c = cv2.imread(r'C:\Users\eebla\OneDrive\Documents\cannoli.jpg')
plt.imshow(cv2.cvtColor(img_c,cv2.COLOR_BGR2RGB))
plt.show()

#Biscotti
img_b= cv2.imread(r'C:\Users\eebla\OneDrive\Documents\biscotti.jpg')
plt.imshow(cv2.cvtColor(img_b,cv2.COLOR_BGR2RGB))
plt.show()
#Biscotti 2
img_b2= cv2.imread(r'C:\Users\eebla\OneDrive\Documents\biscotti2.jpg')
plt.imshow(cv2.cvtColor(img_b2,cv2.COLOR_BGR2RGB))
plt.show()

#both my dogs
img_both= cv2.imread(r'C:\Users\eebla\OneDrive\Documents\both.jpg')
plt.imshow(cv2.cvtColor(img_both,cv2.COLOR_BGR2RGB))
plt.show()

#seal teset image
img_p1= cv2.imread(r'C:\Users\eebla\OneDrive\Documents\seal_test.jpg')
plt.imshow(cv2.cvtColor(img_p1,cv2.COLOR_BGR2RGB))
plt.show()

#seal test image
img_p2= cv2.imread(r'C:\Users\eebla\OneDrive\Documents\seal_test_2.jpg')
plt.imshow(cv2.cvtColor(img_p2,cv2.COLOR_BGR2RGB))
plt.show()
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
In [21]:
#we need to resize these images before passing through neural network
resize_c = tf.image.resize(img_c, IMAGE_SIZE)
resize_b = tf.image.resize(img_b, IMAGE_SIZE)
resize_b2 = tf.image.resize(img_b2,IMAGE_SIZE)
resize_both = tf.image.resize(img_both, IMAGE_SIZE)
resize_p1 = tf.image.resize(img_p1, IMAGE_SIZE)
resize_p2= tf.image.resize(img_p2, IMAGE_SIZE)

# checking, it works
#plt.imshow(resize_c.numpy().astype(int))
#plt.show()
No description has been provided for this image
In [35]:
#scale and predict
yhatc = model.predict(np.expand_dims(resize_c/255, 0)) 
yhatb = model.predict(np.expand_dims(resize_b/255, 0))
yhatb2 = model.predict(np.expand_dims(resize_b2/255, 0))
yhatboth = model.predict(np.expand_dims(resize_both/255, 0))
yhatp1 = model.predict(np.expand_dims(resize_p1/255, 0))
yhatp2 = model.predict(np.expand_dims(resize_p2/255, 0))

#function to print out what the predictions are
def what_dog(pred):
    if pred > 0.5: 
        print(f'Predicted class is a Seal: {pred}')
    else:
        print(f'Predicted class is a Dachshund: {pred}')

#got them all correct
what_dog(yhatc) 
what_dog(yhatb) 
what_dog(yhatb2) 
what_dog(yhatboth) 
what_dog(yhatp1) 
what_dog(yhatp2) 

1/1 [==============================] - 1s 1s/step
1/1 [==============================] - 1s 510ms/step
1/1 [==============================] - 1s 834ms/step
1/1 [==============================] - 1s 746ms/step
1/1 [==============================] - 1s 597ms/step
1/1 [==============================] - 1s 923ms/step
Predicted class is a Dachshund: [[0.00379731]]
Predicted class is a Dachshund: [[0.0075904]]
Predicted class is a Dachshund: [[0.09564621]]
Predicted class is a Dachshund: [[0.16695812]]
Predicted class is a Seal: [[0.9739798]]
Predicted class is a Seal: [[0.93229675]]